Merge tag 'perf-core-2020-06-01' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull perf updates from Ingo Molnar:
 "Kernel side changes:

   - Add AMD Fam17h RAPL support

   - Introduce CAP_PERFMON to kernel and user space

   - Add Zhaoxin CPU support

   - Misc fixes and cleanups

  Tooling changes:

   - perf record:

     Introduce '--switch-output-event' to use arbitrary events to be
     setup and read from a side band thread and, when they take place a
     signal be sent to the main 'perf record' thread, reusing the core
     for '--switch-output' to take perf.data snapshots from the ring
     buffer used for '--overwrite', e.g.:

	# perf record --overwrite -e sched:* \
		      --switch-output-event syscalls:*connect* \
		      workload

     will take perf.data.YYYYMMDDHHMMSS snapshots up to around the
     connect syscalls.

     Add '--num-synthesize-threads' option to control degree of
     parallelism of the synthesize_mmap() code which is scanning
     /proc/PID/task/PID/maps and can be time consuming. This mimics
     pre-existing behaviour in 'perf top'.

   - perf bench:

     Add a multi-threaded synthesize benchmark and kallsyms parsing
     benchmark.

   - Intel PT support:

     Stitch LBR records from multiple samples to get deeper backtraces,
     there are caveats, see the csets for details.

     Allow using Intel PT to synthesize callchains for regular events.

     Add support for synthesizing branch stacks for regular events
     (cycles, instructions, etc) from Intel PT data.

  Misc changes:

   - Updated perf vendor events for power9 and Coresight.

   - Add flamegraph.py script via 'perf flamegraph'

   - Misc other changes, fixes and cleanups - see the Git log for details

  Also, since over the last couple of years perf tooling has matured and
  decoupled from the kernel perf changes to a large degree, going
  forward Arnaldo is going to send perf tooling changes via direct pull
  requests"

* tag 'perf-core-2020-06-01' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (163 commits)
  perf/x86/rapl: Add AMD Fam17h RAPL support
  perf/x86/rapl: Make perf_probe_msr() more robust and flexible
  perf/x86/rapl: Flip logic on default events visibility
  perf/x86/rapl: Refactor to share the RAPL code between Intel and AMD CPUs
  perf/x86/rapl: Move RAPL support to common x86 code
  perf/core: Replace zero-length array with flexible-array
  perf/x86: Replace zero-length array with flexible-array
  perf/x86/intel: Add more available bits for OFFCORE_RESPONSE of Intel Tremont
  perf/x86/rapl: Add Ice Lake RAPL support
  perf flamegraph: Use /bin/bash for report and record scripts
  perf cs-etm: Move definition of 'traceid_list' global variable from header file
  libsymbols kallsyms: Move hex2u64 out of header
  libsymbols kallsyms: Parse using io api
  perf bench: Add kallsyms parsing
  perf: cs-etm: Update to build with latest opencsd version.
  perf symbol: Fix kernel symbol address display
  perf inject: Rename perf_evsel__*() operating on 'struct evsel *' to evsel__*()
  perf annotate: Rename perf_evsel__*() operating on 'struct evsel *' to evsel__*()
  perf trace: Rename perf_evsel__*() operating on 'struct evsel *' to evsel__*()
  perf script: Rename perf_evsel__*() operating on 'struct evsel *' to evsel__*()
  ...

7 files changed, 347 insertions, 5 deletions

diff --git a/tools/perf/bench/Build b/tools/perf/bench/Build
index e4e321b6f883..768e408757a0 100644
--- a/tools/perf/bench/Build
+++ b/tools/perf/bench/Build
@@ -6,9 +6,10 @@ perf-y += futex-wake.o
 perf-y += futex-wake-parallel.o
 perf-y += futex-requeue.o
 perf-y += futex-lock-pi.o
-
 perf-y += epoll-wait.o
 perf-y += epoll-ctl.o
+perf-y += synthesize.o
+perf-y += kallsyms-parse.o
 
 perf-$(CONFIG_X86_64) += mem-memcpy-x86-64-lib.o
 perf-$(CONFIG_X86_64) += mem-memcpy-x86-64-asm.o
diff --git a/tools/perf/bench/bench.h b/tools/perf/bench/bench.h
index 4aa6de1aa67d..61cae4966cae 100644
--- a/tools/perf/bench/bench.h
+++ b/tools/perf/bench/bench.h
@@ -41,9 +41,10 @@ int bench_futex_wake_parallel(int argc, const char **argv);
 int bench_futex_requeue(int argc, const char **argv);
 /* pi futexes */
 int bench_futex_lock_pi(int argc, const char **argv);
-
 int bench_epoll_wait(int argc, const char **argv);
 int bench_epoll_ctl(int argc, const char **argv);
+int bench_synthesize(int argc, const char **argv);
+int bench_kallsyms_parse(int argc, const char **argv);
 
 #define BENCH_FORMAT_DEFAULT_STR	"default"
 #define BENCH_FORMAT_DEFAULT		0
diff --git a/tools/perf/bench/epoll-wait.c b/tools/perf/bench/epoll-wait.c
index f938c585d512..cf797362675b 100644
--- a/tools/perf/bench/epoll-wait.c
+++ b/tools/perf/bench/epoll-wait.c
@@ -519,7 +519,8 @@ int bench_epoll_wait(int argc, const char **argv)
 		qsort(worker, nthreads, sizeof(struct worker), cmpworker);
 
 	for (i = 0; i < nthreads; i++) {
-		unsigned long t = worker[i].ops / bench__runtime.tv_sec;
+		unsigned long t = bench__runtime.tv_sec > 0 ?
+			worker[i].ops / bench__runtime.tv_sec : 0;
 
 		update_stats(&throughput_stats, t);
 
diff --git a/tools/perf/bench/futex-hash.c b/tools/perf/bench/futex-hash.c
index 65eebe06c04d..915bf3da7ce2 100644
--- a/tools/perf/bench/futex-hash.c
+++ b/tools/perf/bench/futex-hash.c
@@ -205,7 +205,8 @@ int bench_futex_hash(int argc, const char **argv)
 	pthread_mutex_destroy(&thread_lock);
 
 	for (i = 0; i < nthreads; i++) {
-		unsigned long t = worker[i].ops / bench__runtime.tv_sec;
+		unsigned long t = bench__runtime.tv_sec > 0 ?
+			worker[i].ops / bench__runtime.tv_sec : 0;
 		update_stats(&throughput_stats, t);
 		if (!silent) {
 			if (nfutexes == 1)
diff --git a/tools/perf/bench/futex-lock-pi.c b/tools/perf/bench/futex-lock-pi.c
index 89fd8f325f38..bb25d8beb3b8 100644
--- a/tools/perf/bench/futex-lock-pi.c
+++ b/tools/perf/bench/futex-lock-pi.c
@@ -211,7 +211,8 @@ int bench_futex_lock_pi(int argc, const char **argv)
 	pthread_mutex_destroy(&thread_lock);
 
 	for (i = 0; i < nthreads; i++) {
-		unsigned long t = worker[i].ops / bench__runtime.tv_sec;
+		unsigned long t = bench__runtime.tv_sec > 0 ?
+			worker[i].ops / bench__runtime.tv_sec : 0;
 
 		update_stats(&throughput_stats, t);
 		if (!silent)
diff --git a/tools/perf/bench/kallsyms-parse.c b/tools/perf/bench/kallsyms-parse.c
new file mode 100644
index 000000000000..2b0d0f980ae9
--- /dev/null
+++ b/tools/perf/bench/kallsyms-parse.c
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Benchmark of /proc/kallsyms parsing.
+ *
+ * Copyright 2020 Google LLC.
+ */
+#include <stdlib.h>
+#include "bench.h"
+#include "../util/stat.h"
+#include <linux/time64.h>
+#include <subcmd/parse-options.h>
+#include <symbol/kallsyms.h>
+
+static unsigned int iterations = 100;
+
+static const struct option options[] = {
+	OPT_UINTEGER('i', "iterations", &iterations,
+		"Number of iterations used to compute average"),
+	OPT_END()
+};
+
+static const char *const bench_usage[] = {
+	"perf bench internals kallsyms-parse <options>",
+	NULL
+};
+
+static int bench_process_symbol(void *arg __maybe_unused,
+				const char *name __maybe_unused,
+				char type __maybe_unused,
+				u64 start __maybe_unused)
+{
+	return 0;
+}
+
+static int do_kallsyms_parse(void)
+{
+	struct timeval start, end, diff;
+	u64 runtime_us;
+	unsigned int i;
+	double time_average, time_stddev;
+	int err;
+	struct stats time_stats;
+
+	init_stats(&time_stats);
+
+	for (i = 0; i < iterations; i++) {
+		gettimeofday(&start, NULL);
+		err = kallsyms__parse("/proc/kallsyms", NULL,
+				bench_process_symbol);
+		if (err)
+			return err;
+
+		gettimeofday(&end, NULL);
+		timersub(&end, &start, &diff);
+		runtime_us = diff.tv_sec * USEC_PER_SEC + diff.tv_usec;
+		update_stats(&time_stats, runtime_us);
+	}
+
+	time_average = avg_stats(&time_stats) / USEC_PER_MSEC;
+	time_stddev = stddev_stats(&time_stats) / USEC_PER_MSEC;
+	printf("  Average kallsyms__parse took: %.3f ms (+- %.3f ms)\n",
+		time_average, time_stddev);
+	return 0;
+}
+
+int bench_kallsyms_parse(int argc, const char **argv)
+{
+	argc = parse_options(argc, argv, options, bench_usage, 0);
+	if (argc) {
+		usage_with_options(bench_usage, options);
+		exit(EXIT_FAILURE);
+	}
+
+	return do_kallsyms_parse();
+}
diff --git a/tools/perf/bench/synthesize.c b/tools/perf/bench/synthesize.c
new file mode 100644
index 000000000000..8d624aea1c5e
--- /dev/null
+++ b/tools/perf/bench/synthesize.c
@@ -0,0 +1,262 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Benchmark synthesis of perf events such as at the start of a 'perf
+ * record'. Synthesis is done on the current process and the 'dummy' event
+ * handlers are invoked that support dump_trace but otherwise do nothing.
+ *
+ * Copyright 2019 Google LLC.
+ */
+#include <stdio.h>
+#include "bench.h"
+#include "../util/debug.h"
+#include "../util/session.h"
+#include "../util/stat.h"
+#include "../util/synthetic-events.h"
+#include "../util/target.h"
+#include "../util/thread_map.h"
+#include "../util/tool.h"
+#include "../util/util.h"
+#include <linux/atomic.h>
+#include <linux/err.h>
+#include <linux/time64.h>
+#include <subcmd/parse-options.h>
+
+static unsigned int min_threads = 1;
+static unsigned int max_threads = UINT_MAX;
+static unsigned int single_iterations = 10000;
+static unsigned int multi_iterations = 10;
+static bool run_st;
+static bool run_mt;
+
+static const struct option options[] = {
+	OPT_BOOLEAN('s', "st", &run_st, "Run single threaded benchmark"),
+	OPT_BOOLEAN('t', "mt", &run_mt, "Run multi-threaded benchmark"),
+	OPT_UINTEGER('m', "min-threads", &min_threads,
+		"Minimum number of threads in multithreaded bench"),
+	OPT_UINTEGER('M', "max-threads", &max_threads,
+		"Maximum number of threads in multithreaded bench"),
+	OPT_UINTEGER('i', "single-iterations", &single_iterations,
+		"Number of iterations used to compute single-threaded average"),
+	OPT_UINTEGER('I', "multi-iterations", &multi_iterations,
+		"Number of iterations used to compute multi-threaded average"),
+	OPT_END()
+};
+
+static const char *const bench_usage[] = {
+	"perf bench internals synthesize <options>",
+	NULL
+};
+
+static atomic_t event_count;
+
+static int process_synthesized_event(struct perf_tool *tool __maybe_unused,
+				     union perf_event *event __maybe_unused,
+				     struct perf_sample *sample __maybe_unused,
+				     struct machine *machine __maybe_unused)
+{
+	atomic_inc(&event_count);
+	return 0;
+}
+
+static int do_run_single_threaded(struct perf_session *session,
+				struct perf_thread_map *threads,
+				struct target *target, bool data_mmap)
+{
+	const unsigned int nr_threads_synthesize = 1;
+	struct timeval start, end, diff;
+	u64 runtime_us;
+	unsigned int i;
+	double time_average, time_stddev, event_average, event_stddev;
+	int err;
+	struct stats time_stats, event_stats;
+
+	init_stats(&time_stats);
+	init_stats(&event_stats);
+
+	for (i = 0; i < single_iterations; i++) {
+		atomic_set(&event_count, 0);
+		gettimeofday(&start, NULL);
+		err = __machine__synthesize_threads(&session->machines.host,
+						NULL,
+						target, threads,
+						process_synthesized_event,
+						data_mmap,
+						nr_threads_synthesize);
+		if (err)
+			return err;
+
+		gettimeofday(&end, NULL);
+		timersub(&end, &start, &diff);
+		runtime_us = diff.tv_sec * USEC_PER_SEC + diff.tv_usec;
+		update_stats(&time_stats, runtime_us);
+		update_stats(&event_stats, atomic_read(&event_count));
+	}
+
+	time_average = avg_stats(&time_stats);
+	time_stddev = stddev_stats(&time_stats);
+	printf("  Average %ssynthesis took: %.3f usec (+- %.3f usec)\n",
+		data_mmap ? "data " : "", time_average, time_stddev);
+
+	event_average = avg_stats(&event_stats);
+	event_stddev = stddev_stats(&event_stats);
+	printf("  Average num. events: %.3f (+- %.3f)\n",
+		event_average, event_stddev);
+
+	printf("  Average time per event %.3f usec\n",
+		time_average / event_average);
+	return 0;
+}
+
+static int run_single_threaded(void)
+{
+	struct perf_session *session;
+	struct target target = {
+		.pid = "self",
+	};
+	struct perf_thread_map *threads;
+	int err;
+
+	perf_set_singlethreaded();
+	session = perf_session__new(NULL, false, NULL);
+	if (IS_ERR(session)) {
+		pr_err("Session creation failed.\n");
+		return PTR_ERR(session);
+	}
+	threads = thread_map__new_by_pid(getpid());
+	if (!threads) {
+		pr_err("Thread map creation failed.\n");
+		err = -ENOMEM;
+		goto err_out;
+	}
+
+	puts(
+"Computing performance of single threaded perf event synthesis by\n"
+"synthesizing events on the perf process itself:");
+
+	err = do_run_single_threaded(session, threads, &target, false);
+	if (err)
+		goto err_out;
+
+	err = do_run_single_threaded(session, threads, &target, true);
+
+err_out:
+	if (threads)
+		perf_thread_map__put(threads);
+
+	perf_session__delete(session);
+	return err;
+}
+
+static int do_run_multi_threaded(struct target *target,
+				unsigned int nr_threads_synthesize)
+{
+	struct timeval start, end, diff;
+	u64 runtime_us;
+	unsigned int i;
+	double time_average, time_stddev, event_average, event_stddev;
+	int err;
+	struct stats time_stats, event_stats;
+	struct perf_session *session;
+
+	init_stats(&time_stats);
+	init_stats(&event_stats);
+	for (i = 0; i < multi_iterations; i++) {
+		session = perf_session__new(NULL, false, NULL);
+		if (!session)
+			return -ENOMEM;
+
+		atomic_set(&event_count, 0);
+		gettimeofday(&start, NULL);
+		err = __machine__synthesize_threads(&session->machines.host,
+						NULL,
+						target, NULL,
+						process_synthesized_event,
+						false,
+						nr_threads_synthesize);
+		if (err) {
+			perf_session__delete(session);
+			return err;
+		}
+
+		gettimeofday(&end, NULL);
+		timersub(&end, &start, &diff);
+		runtime_us = diff.tv_sec * USEC_PER_SEC + diff.tv_usec;
+		update_stats(&time_stats, runtime_us);
+		update_stats(&event_stats, atomic_read(&event_count));
+		perf_session__delete(session);
+	}
+
+	time_average = avg_stats(&time_stats);
+	time_stddev = stddev_stats(&time_stats);
+	printf("    Average synthesis took: %.3f usec (+- %.3f usec)\n",
+		time_average, time_stddev);
+
+	event_average = avg_stats(&event_stats);
+	event_stddev = stddev_stats(&event_stats);
+	printf("    Average num. events: %.3f (+- %.3f)\n",
+		event_average, event_stddev);
+
+	printf("    Average time per event %.3f usec\n",
+		time_average / event_average);
+	return 0;
+}
+
+static int run_multi_threaded(void)
+{
+	struct target target = {
+		.cpu_list = "0"
+	};
+	unsigned int nr_threads_synthesize;
+	int err;
+
+	if (max_threads == UINT_MAX)
+		max_threads = sysconf(_SC_NPROCESSORS_ONLN);
+
+	puts(
+"Computing performance of multi threaded perf event synthesis by\n"
+"synthesizing events on CPU 0:");
+
+	for (nr_threads_synthesize = min_threads;
+	     nr_threads_synthesize <= max_threads;
+	     nr_threads_synthesize++) {
+		if (nr_threads_synthesize == 1)
+			perf_set_singlethreaded();
+		else
+			perf_set_multithreaded();
+
+		printf("  Number of synthesis threads: %u\n",
+			nr_threads_synthesize);
+
+		err = do_run_multi_threaded(&target, nr_threads_synthesize);
+		if (err)
+			return err;
+	}
+	perf_set_singlethreaded();
+	return 0;
+}
+
+int bench_synthesize(int argc, const char **argv)
+{
+	int err = 0;
+
+	argc = parse_options(argc, argv, options, bench_usage, 0);
+	if (argc) {
+		usage_with_options(bench_usage, options);
+		exit(EXIT_FAILURE);
+	}
+
+	/*
+	 * If neither single threaded or multi-threaded are specified, default
+	 * to running just single threaded.
+	 */
+	if (!run_st && !run_mt)
+		run_st = true;
+
+	if (run_st)
+		err = run_single_threaded();
+
+	if (!err && run_mt)
+		err = run_multi_threaded();
+
+	return err;
+}